In order to estimate human risk based on animal carcinogenesis experiments, it is currently necessary to extrapolate data from high dose experiments to low doses. The purpose of this investigation is to study the various mathematical and probabilistic methods for extrapolation with particular emphasis on those which attempt to relate mechanistically to the carcinogenic process.